diff --git a/evm_arithmetization/src/memory/segments.rs b/evm_arithmetization/src/memory/segments.rs
index e1b6678f6..a9534830a 100644
--- a/evm_arithmetization/src/memory/segments.rs
+++ b/evm_arithmetization/src/memory/segments.rs
@@ -100,10 +100,12 @@ impl Segment {
     pub(crate) const COUNT: usize = 39;
 
     /// Unscales this segment by `SEGMENT_SCALING_FACTOR`.
+    #[inline(always)]
     pub(crate) const fn unscale(&self) -> usize {
         *self as usize >> SEGMENT_SCALING_FACTOR
     }
 
+    #[inline(always)]
     pub(crate) const fn all() -> [Self; Self::COUNT] {
         [
             Self::Code,
diff --git a/evm_arithmetization/src/witness/memory.rs b/evm_arithmetization/src/witness/memory.rs
index ff616a348..7abced8a1 100644
--- a/evm_arithmetization/src/witness/memory.rs
+++ b/evm_arithmetization/src/witness/memory.rs
@@ -23,7 +23,8 @@ use crate::witness::errors::ProgramError;
 use crate::witness::errors::ProgramError::MemoryError;
 
 impl MemoryChannel {
-    pub(crate) fn index(&self) -> usize {
+    #[inline(always)]
+    pub(crate) const fn index(&self) -> usize {
         match *self {
             Code => 0,
             GeneralPurpose(n) => {
@@ -43,6 +44,7 @@ pub struct MemoryAddress {
 }
 
 impl MemoryAddress {
+    #[inline(always)]
     pub(crate) const fn new(context: usize, segment: Segment, virt: usize) -> Self {
         Self {
             context,
@@ -69,7 +71,8 @@ impl MemoryAddress {
         Ok(Self::new(context, Segment::all()[segment], virt))
     }
 
-    pub(crate) fn increment(&mut self) {
+    #[inline(always)]
+    pub(crate) const fn increment(&mut self) {
         self.virt = self.virt.saturating_add(1);
     }
 }
@@ -104,7 +107,8 @@ pub(crate) static DUMMY_MEMOP: MemoryOp = MemoryOp {
 };
 
 impl MemoryOp {
-    pub(crate) fn new(
+    #[inline(always)]
+    pub(crate) const fn new(
         channel: MemoryChannel,
         clock: usize,
         address: MemoryAddress,
@@ -123,6 +127,7 @@ impl MemoryOp {
         }
     }
 
+    #[inline(always)]
     pub(crate) const fn new_dummy_read(
         address: MemoryAddress,
         timestamp: usize,
@@ -137,6 +142,7 @@ impl MemoryOp {
         }
     }
 
+    #[inline(always)]
     pub(crate) const fn sorting_key(&self) -> (usize, usize, usize, usize) {
         (
             self.address.context,
@@ -175,6 +181,7 @@ impl MemoryState {
         }
     }
 
+    #[inline]
     pub(crate) fn get(&self, address: MemoryAddress) -> Option<U256> {
         if address.context >= self.contexts.len() {
             return None;
@@ -188,7 +195,7 @@ impl MemoryState {
             return None;
         }
         let val = self.contexts[address.context].segments[address.segment].get(address.virt);
-        assert!(
+        debug_assert!(
             val.bits() <= segment.bit_range(),
             "Value {} exceeds {:?} range of {} bits",
             val,
@@ -245,6 +252,7 @@ impl MemoryState {
         }
     }
 
+    #[inline]
     pub(crate) fn set(&mut self, address: MemoryAddress, val: U256) {
         while address.context >= self.contexts.len() {
             self.contexts.push(MemoryContextState::default());
@@ -252,7 +260,7 @@ impl MemoryState {
 
         let segment = Segment::all()[address.segment];
 
-        assert!(
+        debug_assert!(
             val.bits() <= segment.bit_range(),
             "Value {} exceeds {:?} range of {} bits",
             val,
@@ -320,6 +328,7 @@ pub(crate) struct MemorySegmentState {
 }
 
 impl MemorySegmentState {
+    #[inline]
     pub(crate) fn get(&self, virtual_addr: usize) -> U256 {
         self.content
             .get(virtual_addr)
@@ -328,6 +337,7 @@ impl MemorySegmentState {
             .unwrap_or_default()
     }
 
+    #[inline]
     pub(crate) fn set(&mut self, virtual_addr: usize, value: U256) {
         if virtual_addr >= self.content.len() {
             self.content.resize(virtual_addr + 1, None);
diff --git a/evm_arithmetization/src/witness/traces.rs b/evm_arithmetization/src/witness/traces.rs
index 3ff68d8b6..cb9e605e8 100644
--- a/evm_arithmetization/src/witness/traces.rs
+++ b/evm_arithmetization/src/witness/traces.rs
@@ -126,10 +126,12 @@ impl<T: Copy + Field> Traces<T> {
         self.poseidon_ops.truncate(checkpoint.poseidon_len);
     }
 
+    #[inline(always)]
     pub(crate) fn mem_ops_since(&self, checkpoint: TraceCheckpoint) -> &[MemoryOp] {
         &self.memory_ops[checkpoint.memory_len..]
     }
 
+    #[inline(always)]
     pub(crate) fn clock(&self) -> usize {
         self.cpu.len()
     }
diff --git a/evm_arithmetization/src/witness/util.rs b/evm_arithmetization/src/witness/util.rs
index 5bd103096..b08009e06 100644
--- a/evm_arithmetization/src/witness/util.rs
+++ b/evm_arithmetization/src/witness/util.rs
@@ -16,12 +16,14 @@ use crate::memory::segments::Segment;
 use crate::witness::errors::ProgramError;
 use crate::witness::memory::{MemoryAddress, MemoryChannel, MemoryOp, MemoryOpKind};
 
+#[inline(always)]
 fn to_byte_checked(n: U256) -> u8 {
     let res = n.byte(0);
     assert_eq!(n, res.into());
     res
 }
 
+#[inline(always)]
 fn to_bits_le<F: RichField>(n: u8) -> [F; 8] {
     let mut res = [F::ZERO; 8];
     for (i, bit) in res.iter_mut().enumerate() {
@@ -76,7 +78,8 @@ pub(crate) fn fill_channel_with_value<F: RichField>(
 
 /// Pushes without writing in memory. This happens in opcodes where a push
 /// immediately follows a pop.
-pub(crate) fn push_no_write<F: RichField>(state: &mut GenerationState<F>, val: U256) {
+#[inline(always)]
+pub(crate) const fn push_no_write<F: RichField>(state: &mut GenerationState<F>, val: U256) {
     state.registers.stack_top = val;
     state.registers.stack_len += 1;
 }
@@ -135,6 +138,7 @@ pub(crate) fn mem_read_with_log<F: RichField>(
     (val, op)
 }
 
+#[inline(always)]
 pub(crate) fn mem_write_log<F: RichField>(
     channel: MemoryChannel,
     address: MemoryAddress,